# Audio pipeline in depth

This chapter explains how an uploaded sound file ends up playing for the right
participant. You set the pieces up in {doc}`setup/6-audio`. The examples use the
Acoustic Ripple Therapy scheme (see {doc}`about-the-trial`).

## The building blocks

Sound type
: The category of modulated sound, taken from the participant's current arm. ART uses
  **Amp** and **Phase** (with Amp+Phase and Loc planned).

Hearing profile
: A preset describing the participant's hearing. ART uses four: **None**, **Mild**,
  **Mod**, **Sev**.

Modulation band
: The frequency band a sound is modulated in. ART currently uses seven bands ranging
  from 1 to 2 kHz up to 8 to 16 kHz. An *unmodulated* sound (for a washout-unmodulated
  arm) has no band.

## Matching a participant, step by step

Before treatment, the participant is personalised through a short in-app process:

1. **Hearing profile.** They listen to four pure-tone sequences (`PT_None`, `PT_Mild`,
   `PT_Mod`, `PT_Sev`) and pick the one with the most consistent volume. That choice
   sets their hearing profile.
2. **Tinnitus type.** They compare a pure-tone (PT) and a narrowband-noise (NBN)
   sequence for their profile and pick whichever is more like their tinnitus. This
   feeds frequency matching but not the final sound choice.
3. **Frequency match.** They adjust a frequency scale to find the tone closest to their
   tinnitus. The matching sounds are named `{Type}_{Hearing}_{Freq}`, for example
   `PT_Mild_2.8`; the participant only sees the frequency, because type and hearing are
   already set.

## From the match to the sound they hear

The trial's **assignment table** turns the participant's matched tinnitus frequency into
the modulation band they receive. Each frequency has an **active** band and a **placebo**
band; which column applies is set per arm (active arms read the active column, sham arms
read the placebo column).

The therapeutic sound is then the file matching three things: the arm's **sound type**,
the participant's **hearing profile**, and that **modulation band**. Files follow the
naming convention:

```
{SoundType}_{HearingProfile}_{ModBand}
```

for example `Amp_Mod_2-4`. Unmodulated sounds have no band.

### Example assignment table

The table below is the example ART assignment. A participant matched at 2.8 kHz gets a
`2-4` band on the active arm and `4-8` on the sham arm.

| Tinnitus match (kHz) | Active band (kHz) | Placebo band (kHz) |
| --- | --- | --- |
| 1.0 / 1.2 / 1.4 | 1-2 | 2-4 |
| 1.7 / 2.0 | 1.4-2.8 | 2.8-5.7 |
| 2.4 / 2.8 | 2-4 | 4-8 |
| 3.4 / 4.0 | 2.8-5.7 | 5.7-11 |
| 4.8 / 5.7 | 4-8 | 2-4 |
| 6.7 / 8.0 / 9.5 | 5.7-11 | 2.8-5.7 |
| 11.0 / 13.0 / 16.0 | 8-16 | 4-8 |

The assignment table is a trial-level setting, so a future trial could use a different
mapping.

## Coverage

Because a participant could fall into any combination of hearing profile and modulation
band, {doc}`setup/6-audio` shows a coverage summary per sound type: how many of the
expected hearing-profile by modulation-band combinations you have uploaded, so gaps are
visible before launch. Sound files are around 60 minutes each; a lossless format is
preferred where space allows.